AI training

[edanko]

There are certain visually distinctive genera and families of Diptera that cannot be confidently identified further without close examination of a specimen, except for a handful of species which are themselves visually distinctive in a differing way. A few examples include Pollenia spp., Sciaridae, and the greater part of Sarcophagidae.

Because the iNat AI only trains on research grade observations, the AI effectively doesn't know that these taxa "exist." If you try to use the AI on one of them, it will typically suggest species that are related, but will completely omit these groups. This is despite the aforementioned groups being among the most abundantly observed Diptera on iNaturalist.

I don't think there would be any way to include these in the AI without training the AI on community IDed but non-research-grade observations, which has been discussed. Is there any thought of implementing this in the future?

[Scott Loarie]

While its true that the computer vision only works on species at the
moment, this has more to do with the fact that the model can only
accommodate a flat set of classes (e.g. species) and not a hierarchy
nested classes (e.g. family, genus, species etc.) more than it has to
do with the fact that mainly species have are 'Research Quality'.

We're currently actively working on making the computer vision model
work on the full taxonomy and not just species (which includes making
use of data beyond just the 'Research Quality' subset). Attached is a
screenshot from iNat research collaborator Grant Van Horn's work
(https://www.inaturalist.org/people/gvanhorn) which shows a demo of
computer vision working across the taxonomy. Its confidence that this
drawing is a bird is 0.99, that its in the woodpecker order is 0.92,
and that its Pilleated Woodpecker is 0.75.

A lot of the work I've personally been doing these last few weeks with
taxonomy related coordination and functionality has been towards this
goal of making sure iNat's taxonomy can be actively curated but is
also well structured enough to work with this new model. So in case
anyone's wondering its not just because I'm totally OCD about taxonomy
;)

> --
> You received this message because you are subscribed to the Google Groups "iNaturalist" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to inaturalist...@googlegroups.com.
> To post to this group, send email to inatu...@googlegroups.com.
> Visit this group at https://groups.google.com/group/inaturalist.
> For more options, visit https://groups.google.com/d/optout.



--
--------------------------------------------------
Scott R. Loarie, Ph.D.
Co-director, iNaturalist.org
California Academy of Sciences
55 Music Concourse Dr
San Francisco, CA 94118
--------------------------------------------------